Forecasting Model Comparison for Soil Moisture to Obtain Optimal Plant Growth

The advent of industry 4.0 has seen a massive increase in the connectivity of electronic devices to the internet. It also results in the implementation of data gathering schemes for environmental factors. One such field is the agricultural sector. In Sri Lanka, where this research was conducted, agriculture accounts for one fifth of the country’s gross national production. The introduction of wireless sensor networks in the field of agriculture has shown some of the underlying factors that affect the crops and by extension, the harvest and, yields. Recoding of environmental factors such as soil moisture, temperature, humidity, sunlight, etc. has enabled the modeling of the conditions in the plantations and nurseries. Thereby, delivering an understanding of what suboptimized factors can be improved. Also, two models are utilized to forecast the next-step moisture content at Boralanda town in Sri Lanka based on previous read values: Seasonal Autoregressive Integrated Moving Average (SARIMA) and Long-Short Term Memory (LSTM) Neural Network. It is shown that the LSTM model is superior with much lower error when predicting many time steps.